Is High-Res Music ‘Dead’?

Music
195

Sorry to jump in this discussion, but I had the same doubt and wanted to check if I can hear any difference in my room with my components. Here is what I did (and the results). I’d like your opinion on my approach (if you have time and willingness to review it, of course).

  1. I got one on my LP and choose a track of almost 5 minutes (“Cold, Dark and Yesterday” from an Hall&Oates EP of my youth) to test with. Maybe not the best track to test with, so if you have better suggestion let me know and I’ll see to repeat the test.

  2. Recorded that track at 96/32 with the following chain:

TT → Ampli Phono In → Ampli Rec Out → SB Play3 Mic in → Audacity

I suppose I was all analog up to the ADC, although I’m not sure how the EP was originally recorded. However that shouldn’t impact the result of the test, because I would be listening to a recording of the same analog signal out of the ampli.

  1. Saved the track at 96/24

  2. Copied the track to another file; downsampled to 44/16 and saved it

  3. Copied the 96/32 track to another file and downsampled it to 8/16 and saved it.

At this point, I have the same track at three different resolution 96/24, 44/16 and 8/16. I made the track 8/16 to verify that I can effectively distinguish a “worst” SQ, although I could not be able to differentiate between “good” ones.

I could play these tracks randomly, and check. But I was not satisfied yet, so I made the following:

  1. split each track in 1 minute sections and then mixed High/Low together (upsampling the low quality to 96/24). This way, I ended up with two files:

file1 (96/24 + 44/16 mix) = 1 min at 96/24 + 1 min at 44/16 + 1min 96/24 + 1min 44/16 + 1min 96/24
file2 (96/24 + 8/16 mix) = 1 min at 96/24 + 1min at 8/16 + 1min 96/24 + 1min 8/16 + 1min 96/24

My idea was to check the effect of the “upsampling” and see if it could make any difference.

  1. Finally, I placed the 1 minute tracks one after the other in a playlist, keeping their original resolution. Sort of Step6, but without the upsampling. Here, I wanted to rule out the upsampling and minimize as possible the audio memory effect (I mean: listening to 5 minutes of song at a resolution and then a 5 minutes at another resolution). Also, this had a gap between tracks, giving a cue that something was changing so not completely blind test.

I finally sat down and played the tracks, using Roon. Here my results.

The track at 8/16 is clearly bad. I mean… it hurts for how bad it is. Fine: I can recognize bad if I hear it.

The tracks at 44/16 and 96/24 are the same to my ears. Even knowing what I was playing, I couldn’t find any audible difference.

The mix 96/8 was bad too. I can say without doubt when the 8/16 sections begin. You can see the high freq leaving the room, literally. Fine: I can distinguish good/bad and upsampling does not add any information.

The mix 96/44 is indistinguishable. There is no way FOR ME to hear any difference. It could either have been the 96/24 or the 44/16 track for what I could say.

Next, I borrowed my wife’s ears. She didn’t know anything about my experiment; it was the first time she heard the song; she’s definitively not an audiophile (and doesn’t care); for her 96/24 and 44/16 are just numbers.

So I sat her in the listening chair and moved behind the Roon remote. She couldn’t see what I was playing and all I asked her was “Tell me if you can hear any difference between this tracks. They are all the same song”.

I played the 44/16 first, to set the reference. Next I played the 8/16. She was able immediately to say something was bad. So far so good. Next I played 96/24 and she said it was like the first one. Next I played the 96/44 mix: no reaction, she didn’t hear anything different. When I played the 96/8 bit, again she caught the difference (I mean: a deaf would hear it).

So, she wasn’t able to hear any difference between the 44/16 and the 96/24 when blinded.

Next I tell here that I would play sections of the song at different resolution, and please tell me if you can hear any difference KNOWING THAT THEY ARE different. Still, she didn’t know how they were different, only that they were.

I played the playlist 96/24 + 44/16 (which has stops between each section) without telling her which was which. This time she said “They sound the same to me, but if I really must find a difference, the second and fourth ones (44/16) are a little bit WARMER ( :-o ). I like them more. But I’m saying it only because you tell me that there is a difference, so I’m looking for it. Not knowing it, they are the same to me”.

So, for me in my actual environment with my old ears and with my old gears, 96/24 or 44/16 makes no difference and upsampling something does not improve the quality of the signal at all (which makes sense: if info is not there to start with…).

Anyhow, the fact that my wife (without better knowledge) recognized the 44/16 as warmer (WHEN TOLD TO SEARCH FOR DIFFERENCES!) with respect to the 96/24 make me think that MAYBE something could be there. After all, theory tells that there should be more high frequencies in 96/24 and she really does not like high frequencies. So, I’m now ripping my LP at 96/24, just in case future improvement in the gears would change the listening results. But for now, I’m good with CD quality.

5 Likes
196

Hi, let me start by saying that I really appreciate your taking the time to try this! I think everyone owes it to themselves to do true A/B testing at least once, before deciding what sounds better or concluding that it is not necessary or is flawed beyond repair. I don’t know how you managed to play the tracks randomly or whether you were able to seamlessly switch from one to another during playback or jump back and forth, but that’s the only part that can qualify as double-blind. If you want to go a step further, please read on.

  1. Source: Avoid using digitized vinyl for this, as it could be argued that vinyl doesn’t have the bandwidth and/or dynamic range to qualify as hi-res. There are quite a few studio hi-res tracks you can download for free, both PCM and DSD (e.g. 2L High Resolution Music .:. free TEST BENCH).
  2. Down-sampling: After down-sampling hi-res to 44.1/16 or lower, it is important to always up-sample back to the original resolution before comparison. Not only does that eliminate potential differences in the DAC’s handling of different resolutions, it may also eliminate other subtle clues something has changed. (For example, one of my DACs makes a clicking sound when format changes.)
  3. Using someone else: They should be using the same tools you used by themselves. When you administer the test:
  • Unless you toss a coin to decide which one to play, it won’t be random anymore.
  • It’s not going to be double-blind anymore, since it’s not blind to you, and you may give the subject unconscious clues.
  1. While splicing the tracks makes for an interesting experiment, it’s not random or blind, since alternation is not randomness and the transition points are at known intervals.

When I want to test, I use a program I wrote that goes to some lengths to make sure switching between versions is as smooth as possible:

  • Both tracks must be uncompressed and have the same sample format, sample rate and sample count.
  • Both tracks are fully loaded into memory.
  • Since the two blocks of memory have the same size, I can use a single playback position for both, so switching between tracks comes down to choosing one base memory address or another when the driver asks for the next block.

Switching is so smooth, the only way to test it is to use very different qualities so that the differences are obvious - similar to what you did with 8/16. After all this, I reached the same conclusion: hi-res sounds exactly the same to me before and after going through the CD “bottleneck”.

1 Like
197

I came to the same conclusion by both listening and studying technical facts.
As you know, hi-res is not only about file format but whole chain from recording to listener’s ears.
So in addition, I would test my hearing fq range first. Adults cannot even hear full fq range of CD quality.
Next to check are specs of your audio chain components if they really all meet hi-res requirements. Mainly if your speakers have flat fq response up to 192kHz if you choose 384kHz track - and as you know, there are only few speakers with fq response up to 35kHz.
Then I would analyze the content of chosen track - how much ultrasonic content there really is and how loud it is.
After blind test and feeling you could hear some differences, subtract one track from another in some audio program and check for what remains (it will be very, very low dB which not audible) and if there is any correlation in time where you heard differences.

1 Like
198

Hello Marian,

Thanks for taking the time and apologies for not having replied faster.

I didn’t play randomly, but that was not my intention. My wife, however, didn’t know what I was playing to her; couldn’t see me; didn’t know my purpose. I reckon she was blind enough for my intentions.

Thank you. I didn’t know this possibility. Will definitively download them and try again.

I did this in Audacity. The downsampled track were upsampled again when mixed together. So, the 96/44 track had 1st and 3rd minute at 96/48 and the 2nd minute at 44/16 upsampled to 96/48. All the mix was presented as a single song at 96/48; no cue at all. I knew what was going on, but my wife didn’t.

Thanks again for your suggestions. I will repeat the experiment with the tracks you provided and see if I’ll get different results.

199

Hi there,

I’m back. So, I used the first track from the page suggested by Marian to repeat the experiment.

  1. I took the 192k/24b track and downampled it to 96k, 44k, 22k and 11k
  2. I upsampled each track back to 192k
  3. I splitted each track in 30 sec pieces
  4. I joined pieces back together to get the full song with the following sampling flow:

192 - 96 - 192 - 44 - 192 - 22 - 192 - 11 - 96 - 44

Then I listened to it using Roon. It’s not blind and I didn’t want it to be. I wanted to hear differences in sound, even if they were only psychological.

Well, the only transition I was able to appreciate was the 192 - 11 - 96. I listened either from the speakers as well as from the headphone.

I was expecting to hear the 192 - 22 - 192 too, but I checked and I cannot hear much at the 14 kHz range, so that could explain a lot. My ear are gone for good!

So, I picked my 15 yo daughter and asked her to listen and tell me if she could spot any difference. I only told her that I pieced together different resolutions (but not how and where), and to tell me if she could spot any difference.

She listened and spotted the 192 - 11 - 96 immediately (she said “it is like your head goes underwater”). Good, but again no 192 - 22 - 192! So I had her to listen again to that minute and this time (listening with attention) she spotted the transition exactly. So, it is there; you can identify if you have the ears. Good.

But she wasn’t able to tell the 192/96/44 apart. So I had her listen to the 96k version; then to the 44k version. Then I asked which one she preferred. And she said the 44k.

That’s it for me. Definitively for my ears 44/16 is more than enough (at least with the current kit in my environment). I will repeat the experiment in a while, when I will hopefully have better equipment. But for now, I am one of those lucky guys who can forget about hardware and happily spend money for the MUSIC.

Thank to everyone and especially to Marian.

Antonello

7 Likes
200

I’ve invested in a lot of hi-res files, Astell&Kern players, some powerful IEMs and there is absolutely, categorically, no comparison to a well produced, well mixed vinyl (and sometimes CD although only from the 80s or early 90s). Hi-Res is a rip, as the nuances between 16 and 24bit are difficult to distinguish. It does depend on the genre - Tom Petty, Steven Wilson ‘hi-res’ for example is strong, full of dynamic range. And Roon helps to add brightness, but more often than not, hi-res files are very underwhelming.

201

I like that line. :laughing:

202

I’d seen that Darko video. Hi - res was never “mainstream,” simply because record labels have no interest in attaining or maintaining any sort of industry standard. They wanted it to be LPs until it became burdensome to educate buyers about LP playback and maintenance.

And then they wanted it to be CDs until a lower cost alternative (lossy files) were deemed good enough by — once again — a public that wasn’t used to anything better than cassettes played from a Walkman.

That generation makes today’s music. They kinda suck at it, often.

1 Like
203

I am in the camp that thinks Hi-Res music is a bit of a marketing gimmick and a waste of bandwidth.
Didn’t used to think that way, but having listened to various different resolutions I can’t really tell the difference between 16/44 and higher resolutions. I can tell the difference between lossy mp3 and CD quality (but not all the time though).

The marketing element is very clever. Every new DAC / streamer these days seems to be shooting for the stars and majoring on how high their resolution goes - even seeing DSD1024 now!!! , when in reality quality components, design, manufacturing, coding and good QC matter 1000 times more.

I might be totally wrong of course, some people might be able to hear it clearly, but here is the kicker… even if it’s just perception and they really can’t hear a difference in a blind test, that’s perfectly ok because even if you just think hi res sounds better it actually can lead to more enjoyment - more power to you.

9 Likes
204

Agree with to each her own. I’ve quit purchasing the hi-res studio masters from Hyperion Records. The 16/44 versions allow me to greatly enjoy my favorite artist’s performances without compromise. Chain is Zenith SE->Hugo M-scaler->Chord Dave->Focal Utopia headphones. I’m hearing music that, not infrequently, is better than live performance in quality, if not in experience. In my limited experience, the sound engineering quality plays a larger than expected role in my enjoyment. Would have thought this would be routinely excellent, but it definitely is not, particularly due to dynamic range compression, microphone placement/microphones used and venue characteristics.

205

I’m buying more CDs than ever don’t really worry about hires. Most don’t sound that much better than the cd version of the same album, I have switched all my versions as primary instead of the Qobuz hires versions to as I want to play my versions. When I buy downloads it can be any res as mostly but via Bandcamp where you get what your given. Comparing some recent downloads that are cd quality to supposed hires versions on Qobuz I can’t hear a difference.

3 Likes
206

I can’t either

207

Well I use a Project Evolution 9, with GoldRing Ethos, connected with Cardas wires in tone arm and out of base of Arm to my Gold Note PH-10 with it power supply connected with XLR to my Bryton BR-20 to my 5b ST THX (xlr) via Paradigm Studio Monitor 100v3. Yah most records sound better but if you are going to nit pick about a little groove noise then vinyl is not for you.
With this system its a see saw some stuff sounds better digital some better on vinyl. As mentioned its all about gear and what you accept from your sources.

208

Archimago offered his Internet Blind Test: 24-Bit vs. Dithered 16-Bit Part Deux - Daft Punk Edition! And the honest desire to seek truth as audiophiles. this March through June for everyone to test oneself.

Using free Foobar with ABX plugin, I was able to blindly distinguish the 16/24 bit versions, but just barely so.
For me, this put to rest the nagging question whether it‘s worth to pay for Qobuz‘s higher res tier.

It’s most entertaining and enlightening to read the test results - I‘d really love to see the results for those, advocating fancy networking, server PC, streamer, power supply, cabling, etc solutions, while waxing poetic about blacker blacks, sound staging, instrument separation…

1 Like
209

I’m not saying there was cheating (and the results are what I would expect), but if you convert 16 bits to 24 by zero padding and then add random noise in bit zero (LSB), you’d still be able to tell which file was 16 bits, since all samples will have bits 1-7 zeroed out. Internet tests can never be made 100% blind.